Israel-Hamas Conflict, 1 Year Later

  • Background: 10/7/2023 Hamas attacks Israel killing 1,200 people, Israel retaliates. Estimated dead ~44,000.

  • Problem: Hamas Ministry of Health (MoH) reports casualties, cannot be corroborated.

  • Claim: “Women & children disproportionately killed” UN-OHCHR

  • Research Question: To what extent can open-source data be used to identify patterns in the targeting of Palestinian civilians in Gaza?

Methodology: Analyzing Casualty Incidents

Data Source: Airwars tracks civilian incidents from conflicts.

  1. Web Scrape: ~ 800 incidents (~9,000 deaths, ~10,000 injured) store in SQLite.

  2. JSON Parse: Incident characteristics & casualties. Extract geocoordinates from “Assessment” (65% of incidents).

  3. Reverse Geocoding: Submit incident coordinate queries to the Nominatim API, keep location type (i.e., school, hospital).

  4. Sentiment Analysis: Derive emotional tone from assessments. DistilRoBERTa-base, classifies text into Ekman’s 6 basic emotions.

  5. Clustering Analysis: Explore geographic w/ sentiment features to unpack geographic associations with child & women casualties.

Results: Casualty Rate, Sentiment Analysis

  • Children & Women likely to be killed after adjusting for population size.
  • Sadness is the prominent emotional tone, followed by fear, varies across time.

Results: Hierarchical Clustering

  • Cluster 3 (39%), higher average of children killed associated with anger, disgust, fear, number of injured, & occurring in refugee camps.